The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system commands to manage the repository, including package managers (npm, pnpm, yarn, bun, pip, cargo, go) and specialized cleanup tools (knip, ts-prune, vulture). These operations are aligned with the skill's primary function of maintenance and housekeeping.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted content from the repository it is cleaning.
Ingestion points: Reads repository manifests (package.json, pyproject.toml), source code entry points, CI/CD configuration files, and directory structures from the target repository.
Boundary markers: None identified in the prompt templates.
Capability inventory: File read/write, shell command execution (via package managers), and external web search (firecrawl).
Sanitization: Instructions emphasize manual review and separate commits for each phase, providing human-in-the-loop validation, but do not specify technical sanitization of ingested content.
[EXTERNAL_DOWNLOADS]: Fetches package updates from official registries (npm, PyPI, Crates.io, Go Proxy) and uses the firecrawl tool to search for official migration guides and best practices. These sources are considered well-known and appropriate for the task.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill includes a security audit phase that scans for accidentally committed sensitive files (e.g., .env.local, .env.production). It instructs the agent to delete these files and recommend secret rotation, which acts as a security feature rather than an exfiltration risk.

workflow-housekeep